Direct questions

20. How can I tell whether my child is learning or just producing impressive AI projects?

Look for increasing independence on small, relevant tasks—not only the quality of the finished project. Our proposed test is to ask the learner to explain, change, debug and later reuse a concept, while recording what help was needed.

Explore the comparison in this briefing

Activity

01 / 4 options

Explain

Friendly prompt
“Show me how information moves through this part.”
What to observe
Can they connect the steps in their own words or diagram?

Compare each option against the same criteria. The complete table and qualifications remain in the briefing.

The distinction matters empirically. A published high-school mathematics experiment found that performance with AI assistance did not necessarily translate into performance after assistance was removed. That study was about mathematics, not private coding mentorship; it supports caution about confusing assisted output with learning, not a predicted effect size for your child. (pubmed.ncbi.nlm.nih.gov)

What should we ask the learner to do?

Activity Friendly prompt What to observe
Explain “Show me how information moves through this part.” Can they connect the steps in their own words or diagram?
Change “What would need to change if we added another category?” Can they identify the relevant component?
Debug “This example behaves differently. What would you check first?” Can they propose a reason and a test?
Transfer “Can you use the same idea in this smaller new example?” Can they apply the concept outside the rehearsed demo?

Use a simple assistance record: independent / conceptual hint / step-by-step help / completed by someone else. Do not turn it into a score about the child’s worth or intelligence.

Revisit one concept at the next session. Permit normal reference materials and accessibility supports; if a short task is intended to check independent reasoning, agree which generative assistance will be paused.

Finally, keep a contribution record: learner decisions, mentor scaffolding, AI-generated material and third-party components. The objective is honest attribution, not proving that every line was typed unaided.

Sources cited on this page

  1. Generative AI without guardrails can harm learning: Evidence from high school mathematics - PubMed ↗pubmed.ncbi.nlm.nih.gov